Variable attainment among late L2 learners and native speakers: Conditioning factors beyond individual differences

It is widely acknowledged that the outcomes of both first-language (L1) and later-acquired second-language (L2) are variable (e.g. Birdsong, 2018; Dabrowska, 2019). At the same, many studies of grammatical knowledge have revealed a greater degree of variability among L2 learners than among native controls (e.g. DeKeyser, Alfi-Shabtay & Ravid, 2010; Flege, Yeni-Komshian & Liu, 1999; Hartshorne, Tenenbaum & Pinker, 2018). Some researchers (e.g. Andringa, 2014; DeKeyser, 2012) advise against comparisons of natives and learners on items where the performance of natives is variable, in order to pre-empt evidence of overlapping performance ranges.  

                Against this backdrop, this presentation revisits the relationship between language learning and outcome variability. Unlike most preceding work that associates performance variability with individual differences (e.g. deriving from biographical, experiential, conative, identificational, genetic, bio-chemical, aptitude, cognitive style, etc. factors), our focus is on conditioning factors in variable outcomes that are independent of the participants themselves. 

For scalar judgments of sentence acceptability, we compare native and non-native participants' coefficients of variation (CV) at the item level, as a function of: (a) the grammatical status of individual items, i.e. grammatical vs. ungrammatical; (b) the order of item presentation; (c) high vs. low native-speaker rating variability; (d) analytical procedure (here, means analysis vs. correlational analysis). Applying this approach to published data from Birdsong (1992), we find that: (1) CVs of both native French participants and non-natives (Anglophone late learners of French) are significantly larger for ungrammatical items than ungrammatical items, cf. Flege et al. 1999, who found this grammaticality status effect only among learners; (2) CVs of natives and learners alike correlate significantly over all items and on grammatical items, but not on ungrammatical items; in contrast, under means analysis the groups' CVs are not distinguishable as a function of grammatical status; (3) the order of item presentation does not predict CVs for either participant group; (4) learner similarities with natives are unrelated to high/low native-speaker rating variability, cf. DeKeyser (2012). For both groups, we also observe an interaction between the level of native speaker variability and the grammatical status of items. 

This presentation considers two additional participant-independent variables – judgment type (scalar vs. categorical) and the morphosyntactic complexity of items – which may differentially affect variability of ratings, and which contribute to incommensurability of results between otherwise similar studies. 

We offer suggestions about approaches to the study of variability in populations besides natives and learners, and with respect to tasks beyond judgments of acceptability. We go on to propose avenues for future study of the relationship between age of learning and variable outcomes, and we discuss the significance of evidence of similarities in the attained variabilities of natives and learners at relevant levels of observation.